Case 2: Nisha's First Client

How a company of one ran an IFRS to US GAAP job with an AI Worker, and what she would not let it do

Panaversity case study · November 2026

Where this case fits

This is the sixth section of Claude and ChatGPT 2.0, and its second case study. Case 1 showed Nisha managing two AI tools while she sat in front of them. It stayed at the first step of the ladder: a helper used by a person who is present. This case climbs higher. Nisha now owns an AI Worker that runs while she is away, inside limits she sets.

• The brief. One job, start to finish showed the four parts of a brief. Here, the brief becomes the project's standing instructions.
• Work while you sleep. Work that runs while you sleep showed that a scheduled run cannot count on remembering your conversation. Watch what Nisha has to write down so that a run at 1 a.m. can do the job alone.
• A project with threads. Projects: one team lead, many desks showed one conversation handing jobs to threads. Nisha uses one thread for each part of the job, and one to review.
• Saving work somewhere lasting. Why the borrowed desk gets cleared showed that the desk can be cleared. Every version of Nisha's work is saved outside it.
• The second signature. Staying safe showed that a person makes the final decision. Here, the question is who sends anything to the client.
• Identity and the business layer. From personal worker to business worker asked whose name a worker acts under. On Thursday, that question stops being theory.

About this case

This is a fictional teaching case, prepared for class discussion.

Made up for teaching: Nisha, the buyer, the exporter, every figure in the accounts, the fee, the deadline, the emails and Nisha's thoughts. No real company or client data is used.

Not yet recorded: Case 1 was based on a practical exercise that was carried out and recorded. Case 2's practical has not yet been run. The exhibits that would come from running it (the run log, who caught what, and the time and cost) are illustrative, and each one is labeled. The practical run sheet says exactly how to run and record the exercise, so these exhibits can be replaced with real results.¹

Not accounting advice. The accounting differences are real and well known, but the bridge in this case is a short teaching example of selected items. It is not a full conversion and not audit-ready. It needs a qualified accountant's check before anyone relies on it.

A discussion case, not a tested demonstration. Until its practical is run and recorded, Case 2 is a fictional case for discussion. It is not a tested demonstration of an AI Worker. Read what the worker "caught" or "missed" as what might happen, not as findings.

The case follows the style of a Harvard Business School case but is not connected to Harvard. It does not say whether Nisha's choices were right or wrong.

Doing the lab? Here is the order. Read A paid job and Building the AI Worker. Then do the lab. Then read Three nights of work, Thursday night and the exhibits. The later pages show what Nisha found, so reading them first spoils your count of who caught what.

How this case is organized

The story comes first, in four parts. Read them in order.

1. A paid job: Nisha's Monday choice, the client, the job and the brief.
2. Building the AI Worker: the project, the threads, the nightly run, and what the worker can and cannot reach.
3. Three nights of work: what the worker did while she slept, and what she checked each morning.
4. Thursday night: the client asks for more, and Nisha must decide before she goes to bed.

Then come the questions for discussion and the exhibits. For instructors, the teaching note gives a discussion plan. To try the job yourself, the lab walks you through it with a ready-made data pack. The practical run sheet says how to run the exercise and record it. Last, the final assessment gives you a new job to plan on your own. Check yourself ends the case with flashcards and a quiz.